On Wed, Nov 17, 2004 at 07:45:21AM -0800, David Boreham wrote: > Why wouldn't you apply the same divide and conquor strategy > inside a process ? (e.g. by using threads, or multiple descriptor > queues). We're talking about huge systems here. And my feeling here is that any OS on a SMP machine is better at spreading the load across CPUs the better the more separated the different acting parties are. This is especially important for things like cache locality. Threads share memory, and with a NUMA architecture this puts an upper limit on scalability. Volker
Attachment:
pgpQSschnNl3G.pgp
Description: PGP signature